Sparsity-Aware Evolution for Model Merging

TL;DR

This paper introduces a sparsity-aware evolutionary framework for model merging that enhances reliability and sparsity control in large-scale language models through iterative pruning and merging guided by a combined score function.

Contribution

It presents a novel sparsity-aware evolutionary method that integrates pruning into the mutation process and incorporates sparsity constraints into the scoring, improving model merging.

Findings

Improves model merging reliability across large-scale benchmarks.

Effectively promotes sparsity while maintaining performance.

Easy to incorporate into existing workflows due to simplicity.

Abstract

We propose a sparsity-aware evolutionary (SAE) framework for model merging that involves iterative pruning-merging cycles to act as a novel mutation operator. We incorporate the sparsity constraints into the score function, which steers the evolutionary process to favor more sparse models, in addition to other conventional performance scores. Interestingly, the by-product of \textit{competition} for sparsity introduces an extra local \textit{attraction} and interplay into the evolutionary process: if one competitor has more zero elements, the other competitor's non-zero elements will occupy those positions, even though the less sparse competitor loses to the more sparse competitor in other positions. The proposed pipeline is evaluated on a variety of large-scale LLM benchmarks. Experiments demonstrate that our approach can improve model merging reliability across multiple benchmarks, and is easy to incorporate due to its simplicity and being orthogonal to most existing approaches.